import pandas as pd 
import numpy as np 
import matplotlib.pyplot as plt
import seaborn as sns
from warnings import filterwarnings
%matplotlib inline

df = pd.read_csv("Marketing Campaign Performance - data.csv")
df

	country	channel	impressions	clicks	cost	installs	visits	purchases	revenue
0	A	APP-Acquisition-Android	14472269	74022	13,583	12,094	NaN	15226	1839.97
1	A	APP-Acquisition-Android	3057156	27797	4,967	5,403	NaN	4370	763.91
2	A	APP-Acquisition-iOS	537135	8066	4,019	2,394	NaN	1929	749.46
3	A	APP-Acquisition-iOS	518521	7326	3,243	2,100	NaN	1194	452.83
4	A	Web-Retargeting	1851036	8101	1065.45	NaN	7056.0	927	782.33
...	...	...	...	...	...	...	...	...	...
1471	T	SEM-Hotels	1103	18	3.06	NaN	20.0	12	11.27
1472	T	SEM-Hotels	2925	48	2.92	NaN	44.0	1	0.30
1473	T	SEM-Hotels	939	28	2.72	NaN	31.0	4	2.37
1474	T	SEM-Hotels	402	12	1	NaN	12.0	0	NaN
1475	T	SEM-Hotels	375	3	0.22	NaN	3.0	0	NaN

1476 rows × 9 columns

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1476 entries, 0 to 1475
Data columns (total 9 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   country      1476 non-null   object 
 1   channel      1476 non-null   object 
 2   impressions  1476 non-null   int64  
 3   clicks       1476 non-null   int64  
 4   cost         1476 non-null   object 
 5   installs     38 non-null     object 
 6   visits       1428 non-null   float64
 7   purchases    1476 non-null   int64  
 8   revenue      1217 non-null   float64
dtypes: float64(2), int64(3), object(4)
memory usage: 103.9+ KB

#missing values 

df.isna().sum()

country           0
channel           0
impressions       0
clicks            0
cost              0
installs       1438
visits           48
purchases         0
revenue         259
dtype: int64

features_with_na = [features for features in df.columns if df[features].isnull().sum() > 1] #list comprehension to identify null values in Data frame

for feature in features_with_na:
    print(feature, np.round(df[feature].isnull().mean() * 100, 4),  ' % missing values')

installs 97.4255  % missing values
visits 3.252  % missing values
revenue 17.5474  % missing values

As Data is Realtime and no ML model is to be built, so we're avoiding preproprocessing of the data

df.fillna(0) # We won't be deleteing any data, replacing all NA values with 0 

	country	channel	impressions	clicks	cost	installs	visits	purchases	revenue
0	A	APP-Acquisition-Android	14472269	74022	13,583	12,094	0.0	15226	1839.97
1	A	APP-Acquisition-Android	3057156	27797	4,967	5,403	0.0	4370	763.91
2	A	APP-Acquisition-iOS	537135	8066	4,019	2,394	0.0	1929	749.46
3	A	APP-Acquisition-iOS	518521	7326	3,243	2,100	0.0	1194	452.83
4	A	Web-Retargeting	1851036	8101	1065.45	0	7056.0	927	782.33
...	...	...	...	...	...	...	...	...	...
1471	T	SEM-Hotels	1103	18	3.06	0	20.0	12	11.27
1472	T	SEM-Hotels	2925	48	2.92	0	44.0	1	0.30
1473	T	SEM-Hotels	939	28	2.72	0	31.0	4	2.37
1474	T	SEM-Hotels	402	12	1	0	12.0	0	0.00
1475	T	SEM-Hotels	375	3	0.22	0	3.0	0	0.00

1476 rows × 9 columns

. All The Numerical Variables **

numerical_features = [feature for feature in df.columns if df[feature].dtypes != 'O'] # list comprehension feature that are not equal to object type

print('Number of numerical variables: ', len(numerical_features))

# visualise the numerical variables
df[numerical_features].head()

Number of numerical variables:  5

	impressions	clicks	visits	purchases	revenue
0	14472269	74022	NaN	15226	1839.97
1	3057156	27797	NaN	4370	763.91
2	537135	8066	NaN	1929	749.46
3	518521	7326	NaN	1194	452.83
4	1851036	8101	7056.0	927	782.33

# Distribution of the Data

a = 3  # number of rows
b = 2  # number of columns
c = 1  # initialize plot counter

fig = plt.figure(figsize=(30,20))

for i in numerical_features:
    plt.subplot(a, b, c)
    plt.title('{}'.format(i))

    sns.histplot(data= df, x= i)

    c = c + 1

plt.show()

## Numerical variables are usually of 2 type
## 1. Continous variable and Discrete Variables

discrete_feature=[feature for feature in numerical_features if len(df[feature].unique())]
print("Discrete Variables Count: {}".format(len(discrete_feature)))

Discrete Variables Count: 5

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1476 entries, 0 to 1475
Data columns (total 9 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   country      1476 non-null   object 
 1   channel      1476 non-null   object 
 2   impressions  1476 non-null   int64  
 3   clicks       1476 non-null   int64  
 4   cost         1476 non-null   object 
 5   installs     38 non-null     object 
 6   visits       1428 non-null   float64
 7   purchases    1476 non-null   int64  
 8   revenue      1217 non-null   float64
dtypes: float64(2), int64(3), object(4)
memory usage: 103.9+ KB

a = 5  # number of rows
b = 2  # number of columns
c = 1  # initialize plot counter

fig = plt.figure(figsize=(20, 50))
font = {'size': 12}
  
# using rc function
plt.rc('font', **font)

for i in numerical_features:
    plt.subplot(a, b, c)
    plt.title('{} vs Channel '.format(i))
    sns.barplot(x= "channel", y=  i, data= df, palette="deep")
    plt.xticks(rotation = 40)
    c = c + 1

plt.show()

BOXPLOT SHOWS OUTLIERS INCLUDED IN THE DATASET

a = 5  # number of rows
b = 2  # number of columns
c = 1  # initialize plot counter

fig = plt.figure(figsize=(20, 50))
font = {'size': 12}
  
# using rc function
plt.rc('font', **font)

for i in numerical_features:
    plt.subplot(a, b, c)
    plt.title('{} vs Channel '.format(i))
    sns.boxplot(x= "channel", y=  i, data= df, palette="deep")
    plt.xticks(rotation = 40)
    c = c + 1

plt.show()

categorical_features=[feature for feature in df.columns if df[feature].dtypes=='O']
categorical_features

['country', 'channel', 'cost', 'installs']

for feature in categorical_features:
    print('The feature is {} and number of categories are {}'.format(feature,len(df[feature].unique())))

The feature is country and number of categories are 10
The feature is channel and number of categories are 7
The feature is cost and number of categories are 1172
The feature is installs and number of categories are 39

2.1 Which channel’s ads do the users like the most? (hint: click through rate or CTR)

a = 5  # number of rows
b = 3  # number of columns
c = 1  # initialize plot counter

fig = plt.figure(figsize=(20, 50))
font = {'size': 12}
  
# using rc function
plt.rc('font', **font)

for i in numerical_features:
    plt.subplot(a, b, c)
    plt.title('{} vs impressions, SubPlot {}{}{} '.format(i,a, b,c))
    sns.scatterplot(x= "impressions", y=  i, data= df, palette="deep", hue = "channel")
    plt.xticks(rotation = 40)
    c = c + 1

plt.show()

for i in numerical_features:
    g = sns.relplot(data=df, x= "impressions", y=i, col= "channel", col_wrap=6, hue= "channel", height= 4, aspect=.7,kind="scatter", legend= True, palette="deep")
    g.fig.suptitle(f" impressions vs  {i} ", fontweight ="bold")
    plt.subplots_adjust(top=.90)
    plt.xticks(rotation = 90)
plt.show()

focusing on clicks vs impressions graph, Channel SEM brand and APP - Acqusition Android are most liked by the users. As they have high click / impressions.

2.2 What’s the CPI of App channels (APP-Acquisition-Android and APP-Acquisition-iOS) by country and OS?

df["cost"] = pd.to_numeric(df.cost, errors='coerce') #converting cost to numeric data

a = 5  # number of rows
b = 2  # number of columns
c = 1  # initialize plot counter

fig = plt.figure(figsize=(20, 30))
font = {'size': 12}
  
# using rc function
plt.rc('font', **font)

for i in numerical_features[4:]:
    plt.subplot(a, b, c)
    plt.title('{} vs impressions, SubPlot {}{}{} '.format(i,a, b,c))
    sns.scatterplot(x= "cost", y=  i, data= df, palette="deep", hue = "channel")
    plt.xticks(rotation = 40)
    c = c + 1

plt.show()

for i in numerical_features[4:]:
    g = sns.relplot(data=df, x= "cost", y=i, col= "channel", col_wrap=6, hue= "channel", height= 4, aspect=.7,kind="scatter", legend= True, palette="deep")
    g.fig.suptitle(f" cost vs  {i} ", fontweight ="bold")
    plt.subplots_adjust(top=.90)
    plt.xticks(rotation = 90)
plt.show()

df['Result'] = df['revenue']/df['cost']
df['Result'] 

0            NaN
1            NaN
2            NaN
3            NaN
4       0.734272
          ...   
1471    3.683007
1472    0.102740
1473    0.871324
1474         NaN
1475         NaN
Name: Result, Length: 1476, dtype: float64

2.3 In SEM channels, which country has the best conversion rate?

plt.figure(figsize = (30,15))
plt.title('SEM Channles puchases country wise ')
sns.barplot(x= 'country'  , y=  'purchases', data= df, palette="deep", hue = "channel")
plt.xticks(rotation = 0)
plt.show()

from above graph we can conclude that country S has high conversion Rate as it has maxiumum purchases

2.4 Which Web channel has better CPA?

(Cost Per Action or Acquisition) , (CPA) is calculated by dividing the total cost of conversions by the total number of conversions

g = sns.relplot(data=df, x= "cost", y= 'purchases', col= "channel", col_wrap=7, hue= "channel", height= 4, aspect=.7,kind="scatter", legend= True, palette="deep")
g.fig.suptitle(f" Cost Per Action or Acquisition i.e purchase vs cost ", fontweight ="bold")
plt.subplots_adjust(top=.85)
plt.xticks(rotation = 90)
plt.show()

CPA of channel SEM brand is better than other channles as, cost of acqusition is less and purchases by customer is moree.

2.5 Identify the top 3 performing channels of each country in terms of ROI.

The basic ROI calculation is: ROI = (Net Profit/Total Cost)*100.

df.head()

	country	channel	impressions	clicks	cost	installs	visits	purchases	revenue	Result
0	A	APP-Acquisition-Android	14472269	74022	NaN	12,094	NaN	15226	1839.97	NaN
1	A	APP-Acquisition-Android	3057156	27797	NaN	5,403	NaN	4370	763.91	NaN
2	A	APP-Acquisition-iOS	537135	8066	NaN	2,394	NaN	1929	749.46	NaN
3	A	APP-Acquisition-iOS	518521	7326	NaN	2,100	NaN	1194	452.83	NaN
4	A	Web-Retargeting	1851036	8101	1065.45	NaN	7056.0	927	782.33	0.734272

ROI = (df.revenue / df.cost) * 100
data = pd.DataFrame(ROI, columns = ["ROI"])
# Using DataFrame.insert() to add a column
df.insert(10, "ROI", data, True)

plt.figure(figsize=(20,10))
sns.lineplot(data = df, x = "channel", y = "ROI",size = "country" , hue = "country", style = "country", markers=True, dashes=False)
plt.title("ROI of Diffrent channels based on country")
plt.show()

Based on above graph, we can see ROI of all countries are majorly contributed by

1. SEM Brand

2. SEM fligts

3. Web Retargeting

Additional graphs to understand distribution of ROI vs revenue & cost

plt.figure(figsize=(20,10))
sns.lineplot(data = df, x = "ROI", y = "revenue", size = "channel", hue = "channel", palette= "deep", style = "channel", markers=True, dashes=False )
plt.title("ROI vs Revenue")
plt.show()

plt.figure(figsize=(20,10))
sns.lineplot(data = df, x = "ROI", y = "cost",size =  "channel", hue = "channel", style = "channel", markers=True, dashes=False)
plt.title("ROI vs cost")
plt.show()